Research On Uyghur Continuous Speech Recognition System Based On HTK

Speech Recognition change speech data into text sequence, it is the main component of human-computer interaction. With the development of the Speech recognition technology, it becomes non-specific, large vocabulary continuous speech recognition from the initial isolated digital recognition.Uyghur belongs to the Turkic language family of Altaic language system. Uyghur is an agglutinative language. It is possible to produce a very high number of words from the same root with suffixes. Uyghur pronunciation is spliced by a number of phonemes and so have its unique laws on vowel harmony and consonant harmony. Based on Uyghur own characteristics, this paper established a Uyghur continuous speech corpus consisted of 64 speakers'speech data, researched on the selection of the Uyghur continuous speech recognition units . Based on the above study, this paper selected triphone as the basic recognition unit, used Hidden Markov Model tool (HTK) to established a triphone acoustic model, Used many methods such as decision tree, making Tied-State triphones, fixing the Silence Models, increasing Gaussian Mixture Distribution and so on to improve the precision of the models. In the word layer, used Statistics-based Bigram language model, it is suitable for Uyghur voice features.Finally, this paper has done a variety of recognition experiments using test databases based on the built acoustic model and Bigram language model under the DOS environment, the experimental results show that the recognition rate of sentence reached 68.98%, the recognition rate of word achieved 94.65%. Used VC2005 programming environment to do the secondary development based on HTK tools, developed a Uyghur continuous speech recognition system and made real-time speech recognition experiments. The experimental results show that the sentence recognition rate reached 63.31% and 65.67%, the word recognition rate achieved 90.25% and 91.40%, male and female respectively.